Stomp,hawtdispatch,Android: possible jvm bug

[Mario Giammarco]

Hello,
I have done an Android client that uses stomp jms to connect to an ActiveMQ server.
Stomp jms uses hawtdispatch to work.
In a 2.3 Android phone it works. On android 2.3 and 4.0 emulator it works.
On some android 4.0 phones like samsung galaxy note and motorola razr it does not work.
I have supposed a dead lock and I have deeply debugged the app.
I have found no deadlock but a thread waiting on libcore.io.posix.poll.
Looking at the stacktrace it is called by org.fusesource.hawtdispatch.internal.NioManager.

Now I see that NioManager has several workarounds to buggy JRE.

So I think:

- for Android 4.0 there is the need of a more powerful workaround;
- or the right workaround is already in the code but is not activated correctly.

Can you please help me?

Thanks,
Mario

[Christian Posta]

When a thread in the pool isn't servicing events, it will sit in a select() call waiting for IO events as determined by readiness from the OS. There was a bug in the HotSpot VM that is worked around by setting -Dhawtdispatch.workaround-select-spin=true

You can give that a try, but otherwise would be necessary to find out more about what the OS is doing and whether its treating the readiness selectors properly.

--
You received this message because you are subscribed to the Google Groups "hawtdispatch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hawtdispatch...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

--
Christian Posta

http://www.christianposta.com/blog

twitter: @christianposta

[Mario Giammarco]

2013/3/26 Christian Posta <christi...@gmail.com>:

> When a thread in the pool isn't servicing events, it will sit in a select()
> call waiting for IO events as determined by readiness from the OS. There was
> a bug in the HotSpot VM that is worked around by setting
> -Dhawtdispatch.workaround-select-spin=true
>

In android setting properties is difficult. Is there another way to do it?

> You received this message because you are subscribed to a topic in the
> Google Groups "hawtdispatch" group.
> To unsubscribe from this topic, visit
> https://groups.google.com/d/topic/hawtdispatch/GiNkjvdH6lM/unsubscribe?hl=en.
> To unsubscribe from this group and all its topics, send an email to

[Christian Posta]

At the moment, that's the only way. We can cook up a patch, or you can submit one that allows setting the selection strategy another way (or maybe Hiram has some suggestions). 

[Mario Giammarco]

Still debugging. I have modified source code to force the workaround. But now I have discovered that the poll is right when it waits. Infact my app send nothing so it cannot receive a reply.
It sends two packets using hawtdispatch-transport tcp. I do not understand why when I run it with the debugger it sends data. Does the debugger modify java bytecode?